Selection between models through multi-step-ahead forecasting

We develop and show applications of two new test statistics for deciding if one ARIMA model provides significantly better h-step-ahead forecasts than another, as measured by the difference of approximations to their asymptotic mean square forecast errors. The two statistics differ in the variance estimates used for normalization. Both variance estimates are consistent even when the models considered are incorrect. Our main variance estimate is further distinguished by accounting for parameter estimation, while the simpler variance estimate treats parameters as fixed. Their broad consistency properties offer improvements to what are known as tests of Diebold and Mariano (1995) type, which are tests that treat parameters as fixed and use variance estimates that are generally not consistent in our context. We show how these statistics can be calculated for any pair of ARIMA models with the same differencing operator.     

DATA-DRIVEN NONPARAMETRIC SPECTRAL DENSITY ESTIMATORS FOR ECONOMIC TIME SERIES: A MONTE CARLO STUDY

Spectral analysis at frequencies other than zero plays an increasingly important role in econometrics. A number of alternative automated data-driven procedures for nonparametric spectral density estimation have been suggested in the literature, but little is known about their finite-sample accuracy. We compare five such procedures in terms of their mean-squared percentage error across frequencies. Our data generating processes (DGP) include autoregressive-moving average (ARMA) models, fractionally integrated ARMA models and nonparametric models based on 16 commonly used macroeconomic time series. We find that for both quarterly and monthly data the autoregressive sieve estimator is the most reliable method overall.     

Real-Time Forecasting With a Mixed-Frequency VAR

This article develops a vector autoregression (VAR) for time series which are observed at mixed frequencies—quarterly and monthly. The model is cast in state-space form and estimated with Bayesian methods under a Minnesota-style prior. We show how to evaluate the marginal data density to implement a data-driven hyperparameter selection. Using a real-time dataset, we evaluate forecasts from the mixed-frequency VAR and compare them to standard quarterly frequency VAR and to forecasts from MIDAS regressions. We document the extent to which information that becomes available within the quarter improves the forecasts in real time. This article has online supplementary materials.     

An Application of Operational-Subjective Statistical Methods to Rational Expectations

This article presents a sequential scoring analysis of six econometric forecast distributions for the main components of the annual U.S. gross national product (GNP) accounts—nominal GNP, real GNP, and the implicit price deflator. Analysis of sequential forecasts is presented in terms of proper scoring rules. Computations relevant to the calibration and refinement properties of the forecast distributions are discussed. Annual data are studied for the period 1952–1982. The six forecast distributions are distinguished by the different stances they entail with respect to a subjectivist characterization of the rational-expectations hypothesis.     

The Integration of Artificial Neural Networks and Text Mining to Forecast Gold Futures Prices

Although a previous study found that neural network forecasts were more accurate than time series models for predicting Latin American stock indexes, the forecasting accuracy of neural network for predicting gold futures prices has never been discussed. Therefore, the first objective of this study is to compare the forecasting accuracy of a neural network model with that of ARIMA models. Furthermore, the fluctuations in gold futures are not only influenced by the quantitative variables, but also by many nonquantifiable factors, such as wars, international relations, and terrorist attacks. The second objective of this study is therefore to propose the integration of text mining and an artificial neural network to forecast gold futures prices. The historical gold futures prices from 1999 to 2008 were used as training data and testing data, and the prices of 2009 were used to examine the effectiveness of the proposed model. The results of empirical analysis showed that an artificial neural network forecasted gold futures prices better than ARIMA models did. In addition, text mining provided a reasonable explanation of the trend in gold futures prices.     

Producing Monthly Estimates of Unemployment and Employment According to the International Labour Office Definition

Monthly unemployment statistics are available in Britain from a monthly count of the number of people claiming unemployment-related benefits. There has been considerable debate on the appropriateness of this measure. Unemployment and employment statistics are available quarterly from the Labour Force Survey (LFS), using International Labour Office (ILO) definitions. In this paper various options for producing monthly unemployment estimates according to the ILO definition are examined. Methods considered are a monthly LFS, calculating rolling averages from the quarterly LFS, and methods which combine LFS and claimant count data. It is proposed that a monthly LFS of 60 000 households be introduced which can produce monthly estimates of total unemployment and more detailed estimates quarterly. Such a survey would also fill an important gap by providing monthly employment statistics which are needed to provide a complete picture of the labour market.     

Measurement Problems of Inflation Disaggregation

Why do the three quarterly GNP inflation measures differ so much when they are constructed from the same underlying price data? Algebraically and in tables using data of the second quarter of 1984, it is shown that these differences occur because of quarterly shifts in the composition of the nation's product. Disaggregation of the inflation contributions of the GNP components also makes it clear why, for quarterly analyses, the GNP chain price index is superior to both the implicit GNP deflator and the fixed-weight GNP price index. In particular, the implicit GNP deflator can give severely distorted inflation signals.     

Some simulations and applications of forecasting long-memory time-series models

In this paper, we show some results of forecasting based on the ARFIMA(p,d,q) and ARIMA(p,d,q) models. We show, by simulation, that the technique of forecasting of the ARIMA(p,d,q) model can also be used when d is fractional, i.e., for the ARFIMA(p,d,q) model. We also conduct a simulation study to compare the two estimators of d obtained through regression methods. They are used in the hypothesis test to decide whether or not the series has long memory property and are compared on the basis of their k-step ahead forecast errors. The properties of long-memory models are also investigated using an actual set of data.     

Forecasting Study of Shanghai’s and Shenzhen’s Stock Markets Using a Hybrid Forecast Method

In this article, a novel hybrid method to forecast stock price is proposed. This hybrid method is based on wavelet transform, wavelet denoising, linear models (autoregressive integrated moving average (ARIMA) model and exponential smoothing (ES) model), and nonlinear models (BP Neural Network and RBF Neural Network). The wavelet transform provides a set of better-behaved constitutive series than stock series for prediction. Wavelet denoising is used to eliminate some slight random fluctuations of stock series. ARIMA model and ES model are used to forecast the linear component of denoised stock series, and then BP Neural Network and RBF Neural Network are developed as tools for nonlinear pattern recognition to correct the estimation error of the prediction of linear models. The proposed method is examined in the stock market of Shanghai and Shenzhen and the results are compared with some of the most recent stock price forecast methods. The results show that the proposed hybrid method can provide a considerable improvement for the forecasting accuracy. Meanwhile, this proposed method can also be applied to analysis and forecast reliability of products or systems and improve the accuracy of reliability engineering.     

A bootstrap test for time series linearity

A bootstrap algorithm is proposed for testing Gaussianity and linearity in stationary time series, and consistency of the relevant bootstrap approximations is proven rigorously for the first time. Subba Rao and Gabr (1980) and Hinich (1982) have formulated some well-known nonparametric tests for Gaussianity and linearity based on the asymptotic distribution of the normalized bispectrum. The proposed bootstrap procedure gives an alternative way to approximate the finite-sample null distribution of such test statistics. We revisit a modified form of Hinich's test utilizing kernel smoothing, and compare its performance to the bootstrap test on several simulated data sets and two real data sets—the S&P 500 returns and the quarterly US real GNP growth rate. Interestingly, Hinich's test and the proposed bootstrapped version yield substantially different results when testing Gaussianity and linearity of the GNP data.     

基于深度学习LSTM神经网络的全球股票指数预测研究

作为深度学习技术的经典模型之一，长短期记忆（LSTM）神经网络在挖掘序列数据长期依赖关系中极具优势。基于深度神经网络优化技术，本文构造了一个深层LSTM神经网络并将其应用于全球30个股票指数三种不同期限的预测研究，结果发现：①LSTM神经网络具有很强的泛化能力，对全部指数不同期限的预测效果均很稳定；②LSTM神经网络具有优秀的预测精度，相比三种对照模型（SVR，MLP和ARIMA），其对全部指数的平均预测精度在不同期限上均有提升；③LSTM神经网络能够有效控制误差波动，其对全部指数的平均预测稳定度相比三种对照模型在不同期限上亦均有提高。鉴于LSTM神经网络在预测精度和稳定度两方面的优势，其未来在金融预测中将有广阔的应用前景。     

A Bootstrap Test for Circular Data

In this article, variance stabilizing filters are discussed. A new filter with nice properties is proposed which makes use of moving averages and moving standard deviations, the latter smoothed with the Hodrick-Prescott filter. This filter is compared to a GARCH-type filter. An ARIMA model is estimated for the filtered GDP series, and the parameter estimates are used in forecasting the unfiltered series. These forecasts compare well with those of ARIMA, ARFIMA, and GARCH models based on the unfiltered data. The filter does not color white noise.     

A Comparison of Nonparametric Tests of Weak Separability for Annual and Quarterly Data on Consumption,Leisure, and Money

Among the most fundamental assumptions made in economics are utility maximization and the separability of the arguments in the representative consumer's utility function. These assumptions are important for theoretical and empirical applications of economics. In this article, we present results from nonparametric tests of these assumptions of consumer behavior. We find that utility maximization generally obtains with either annual or quarterly per capita data on consumption goods, leisure, and relatively liquid monetary assets. Annual data on consumption goods, leisure, and all monetary assets are consistent with utility maximization. There is some evidence in support of using partial adjustment models when estimating quarterly data models of the demand for monetary assets. Further, annual data on consumption goods and leisure and on liquid monetary assets meet the necessary and sufficient conditions for weak separability. These results support the notion of a monetary aggregate more broadly based than currency plus demand deposits. Separability of monetary assets does not obtain for quarterly data.     

Complete censuses and samples

"Complete decennial censuses are needed for small areas and other domains. Sample surveys yield diverse and timely data. Censuses can also be combined with samples, and sometimes with data from registers, for diverse estimates that are detailed over both space and time, and hence are timely for small domains. Methods of 'postcensal estimates' for small domains are described. We note uses of censuses for improving samples and of samples for improving censuses, and propose a method for cumulating data from 'rolling' (or rotating) periodic (weekly, monthly or quarterly) samples specifically designed to cover the population in detail over designed spans (annual and quinquennial)."     

A Note on Overdifferencing and the Equivalence of Seasonal Time Series Models With Monthly Means and Models With (0, 1, 1)12 Seasonal Parts When ⊖ = 1

Two general models for monthly seasonal time series are considered, one in which seasonality is modeled with monthly means and another in which seasonality is modeled with a (0, 1, 1)₁₂ ARIMA structure. The models are shown to be equivalent if the seasonal moving average parameter (?) is 1 and if the same assumptions about the 12 initial observations are made for both models. The role of the assumptions about the initial observations is analyzed, and it is argued that for practical purposes the two models can be regarded as equivalent when ? = 1. It is observed that the result extends easily to more general models involving overdifferencing.     

Comparison of data analysis procedures for real-time nanoparticle sampling data using classical regression and ARIMA models

Real-time monitoring is necessary for nanoparticle exposure assessment to characterize the exposure profile, but the data produced are autocorrelated. This study was conducted to compare three statistical methods used to analyze data, which constitute autocorrelated time series, and to investigate the effect of averaging time on the reduction of the autocorrelation using field data. First-order autoregressive (AR(1)) and autoregressive-integrated moving average (ARIMA) models are alternative methods that remove autocorrelation. The classical regression method was compared with AR(1) and ARIMA. Three data sets were used. Scanning mobility particle sizer data were used. We compared the results of regression, AR(1), and ARIMA with averaging times of 1, 5, and 10?min. AR(1) and ARIMA models had similar capacities to adjust autocorrelation of real-time data. Because of the non-stationary of real-time monitoring data, the ARIMA was more appropriate. When using the AR(1), transformation into stationary data was necessary. There was no difference with a longer averaging time. This study suggests that the ARIMA model could be used to process real-time monitoring data especially for non-stationary data, and averaging time setting is flexible depending on the data interval required to capture the effects of processes for occupational and environmental nano measurements.     

Real-Time Forecasts of the Real Price of Oil

We construct a monthly real-time dataset consisting of vintages for 1991.1–2010.12 that is suitable for generating forecasts of the real price of oil from a variety of models. We document that revisions of the data typically represent news, and we introduce backcasting and nowcasting techniques to fill gaps in the real-time data. We show that real-time forecasts of the real price of oil can be more accurate than the no-change forecast at horizons up to 1 year. In some cases, real-time mean squared prediction error (MSPE) reductions may be as high as 25% 1 month ahead and 24% 3 months ahead. This result is in striking contrast to related results in the literature for asset prices. In particular, recursive vector autoregressive (VAR) forecasts based on global oil market variables tend to have lower MSPE at short horizons than forecasts based on oil futures prices, forecasts based on autoregressive (AR) and autoregressive moving average (ARMA) models, and the no-change forecast. In addition, these VAR models have consistently higher directional accuracy.     

互联网企业广告收入预测研究——基于低频数据的神经网络和时间序列组合模型

神经网络模型对大样本时间序列的拟合效果优于传统时间序列模型,但对于年度、月度、日度等低频时间序列的预测则难以发挥其优势。鉴于此,本文应用传统时间序列模型和神经网络模型,建立Holtwinters-BP组合模型,利用Holtwinters模型分别拟合各解释变量序列,利用BP模型拟合解释变量和自变量的非线性关系,基于某社交新闻类APP的日广告收入数据进行互联网企业广告收入预测研究。通过与循环神经网络(RNN)模型、长短期记忆神经网络(LSTM)模型等预测结果的对比发现:Holtwinters-BP组合模型的预测精度和稳定性更高;证明多维变量对于广告收入的显著影响,多变量模型的预测准确性高于单变量模型;构建的Holtwinters-BP组合模型对于低频数据预测有较好的有效性和适用性。     

Optimal Forecasts from Markov Switching Models

We derive forecasts for Markov switching models that are optimal in the mean square forecast error (MSFE) sense by means of weighting observations. We provide analytic expressions of the weights conditional on the Markov states and conditional on state probabilities. This allows us to study the effect of uncertainty around states on forecasts. It emerges that, even in large samples, forecasting performance increases substantially when the construction of optimal weights takes uncertainty around states into account. Performance of the optimal weights is shown through simulations and an application to U.S. GNP, where using optimal weights leads to significant reductions in MSFE. Supplementary materials for this article are available online.     

Forecasting multivariate longitudinal binary data with marginal and marginally specified models

Forecasting with longitudinal data has been rarely studied. Most of the available studies are for continuous response and all of them are for univariate response. In this study, we consider forecasting multivariate longitudinal binary data. Five different models including simple ones, univariate and multivariate marginal models, and complex ones, marginally specified models, are studied to forecast such data. Model forecasting abilities are illustrated via a real-life data set and a simulation study. The simulation study includes a model independent data generation to provide a fair environment for model competitions. Independent variables are forecast as well as the dependent ones to mimic the real-life cases best. Several accuracy measures are considered to compare model forecasting abilities. Results show that complex models yield better forecasts.